Counterfactual Confounder Adjustment for Single-cell Differential Expression Analysis
Yongjin Park
Results
Cell type annotation
PsychENCODE’s marker gene selection might be imperfect since some “marker” genes are shared by different cell types. However, it has enough discriminatory power. Using this initial dictionary of marker genes, we can start analyzing 10x data, which consist of these three files:
- Sparse count matrix:
## %%MatrixMarket matrix coordinate integer general
## 17926 70634 119696200
## 56 1 1
## 139 1 1
## 270 1 1
## 324 1 1
## 375 1 1
## 389 1 1
## 420 1 1
## 557 1 1- Cell barcode list:
## AAACGGGAGATCCCGC.1
## AAATGCCTCCAATGGT.1
## AACCATGTCAGTGCAT.1
## AACCATGTCTGTACGA.1
## AACCGCGTCCGCATAA.1
## AACGTTGGTTCAGGCC.1
## AACTGGTGTACCGGCT.1
## AAGACCTAGTTAACGA.1
## AAGGAGCAGCAATCTC.1
## AAGGAGCTCTGCTGCT.1- Gene/feature list (these names should exist in the marker file):
## FO538757.2
## SAMD11
## NOC2L
## KLHL17
## PLEKHN1
## PERM1
## HES4
## ISG15
## AGRN
## RNF223
Basic statistics for cell type annotation
- 1,472 cells per individual: 50.9% (N=745 \(\pm\) 388) cells for Ex, 12.23% (N=179 \(\pm\) 103) cells for In, 3.54% (N= 54 \(\pm\) 35) cells for OPC, 0.33% (N= 4 \(\pm\) 5) cells for Endo, 25.22% (N=380 \(\pm\) 252) cells for Oligo, 2.61% (N= 36 \(\pm\) 25) cells for Microglia, 4.95% (N= 71 \(\pm\) 47) cells for Astro, 0.34% (N= 4 \(\pm\) 4) cells for Per on average.
Differential expression analysis
What if cell types were combined? Cell-type-sorted pseudo-bulk analysis improves statistical power
## Joining, by = "ct"
CoCoA identified 377 genes are differentially-regulated between different AD conditions
- 377 genes (2.72%) consist of 123 found in Ex, 36 found in In, 86 found in Oligo, 58 found in OPC, 24 found in Microglia, 57 found in Astro
## quartz_off_screen
## 2Why do we find more genes with CoCoA?
Confounding effects prevalent…
## Warning: Removed 4702 rows containing non-finite values (stat_smooth).## Warning: Removed 4702 rows containing non-finite values (stat_cor).## Warning: Removed 4702 rows containing missing values (geom_point).
A genomic view combining all the cell types
Genomic views for each cell type
